Quantum, in motion.

How it works

Each trap is a tightly focused laser beam called an optical tweezer. Loading is random: roughly half the traps catch an atom. A camera image shows which traps are full, then a movable tweezer carries spare atoms into the empty sites until the pattern has no gaps. Real machines do this far faster; it is slowed down here.

Why it matters

In a neutral-atom quantum computer, each atom stores a qubit in two of its energy levels, and laser pulses carry out the operations. A missing atom is a missing qubit, so an array without gaps is the starting point. Neutral atoms are one of several competing ways to build qubits. Most companies profiled on this site build other kinds, such as superconducting circuits and trapped ions, though Microsoft's and Amazon's cloud services also give access to partners' neutral-atom machines.

Try it

Rebuild loads the traps at random and sorts them again; every run needs different moves. The readout counts the traps, the atoms caught, the moves and how much of the pattern is filled.

Go deeper: What a qubit is, and what you can read → Trapped ions: IonQ →

How it works

Each row is the particle's probability across the box at one moment: dark where it is unlikely to be found, cream where it is most likely. The particle behaves as a wave. It spreads, bounces off the walls and overlaps itself; where the overlapping parts are in step they reinforce, and where they are out of step they cancel. That interference weaves the pattern, which physicists call a quantum carpet.

Why it matters

At special moments every part of the wave lines up again and the starting shape returns exactly, a quantum revival: the last row repeats the first. Quantum computers rely on the same arithmetic. Their operations add up amplitudes so that some outcomes reinforce and others cancel, and only then does a measurement read one result.

Try it

Weave another starts from a different shape: one packet, two, three, a flat slab, or one packet pushed both ways at once. Each weaves a different rug, and each ends where it began.

Go deeper: Why interference matters →

How it works

The simulation solves the Schrödinger equation for a particle whose average energy is 77% of what it would need to get over the wall. A ball with too little energy always rolls back. A quantum particle's wave does not stop dead at the wall: it fades inside it, and when the wall is thin enough, part of it comes out the other side. The share that gets through is measured from the simulation: 27% at the starting width, and each step thicker cuts it by about 40%.

Why it matters

Tunnelling is built into the superconducting qubits made by IBM, Google and Rigetti. Each one contains at least one Josephson junction: two superconductors separated by a thin insulating layer that pairs of electrons tunnel across. The 2025 Nobel Prize in Physics went to John Clarke, Michel Devoret and John Martinis for experiments in 1984 and 1985 showing that a whole circuit, big enough to be held in the hand, can behave as a single quantum particle: it tunnelled out of a trapped state and had quantised energy levels.

Try it

Make the wall thinner or thicker and watch the share that gets through change. The wall's height stays at 1.3 times the particle's energy.

Go deeper: IBM → Alphabet / Google Quantum AI → Rigetti Computing →

How it works

Each cell is one possible readout of the qubits. Its area is the chance of getting that readout, and its colour stands for its phase, the part of an amplitude that decides how it interferes. Every added qubit splits every cell in two. Each number is a complex amplitude that takes 16 bytes to store, so the memory doubles with every qubit.

Why it matters

This is why ordinary computers struggle to simulate quantum ones by brute force: in 2017 a 45-qubit simulation used half a petabyte of memory on a supercomputer. Smarter methods handle some circuits. But a quantum computer never hands you all 2ⁿ numbers: measuring n qubits gives n bits, one readout. Any advantage has to come from interference, as the next simulation shows.

Try it

Drag the slider, or press Play from one to watch it double. The drawing stops at 13 qubits, 8,192 cells; the count and the memory keep going to 50.

Go deeper: How ordinary computers work → What a qubit is, and what you can read →

How it works

Each bar is the amplitude of one possible answer, and the chance of reading an answer is its bar's height squared. All 64 start equal. Each round has two steps. First the oracle, a check that recognises the answer, flips the sign of its bar. Then every bar is reflected about the average, the dashed line: the flipped bar jumps up and the others shrink a little. After six rounds the answer holds 99.7% of the probability; a seventh would overshoot and lower it to 90.7%.

Why it matters

Checking 64 possibilities one at a time takes about 32 looks on average. Grover's search needs about 0.8 × √N rounds instead of about N/2 looks: roughly 785 rounds for a million possibilities instead of about 500,000. That square-root speedup is real but modest, and researchers have argued that on early error-corrected machines the overheads would cancel it unless error correction becomes far more efficient. Shor's factoring algorithm offers a much larger speedup over the best known classical methods, but needs far larger error-corrected machines. When a headline cites a quantum speedup, ask which kind.

Try it

Tap or click any bar to hide the answer there and search again. The ticker logs every round.

Go deeper: How to read a quantum claim → Optimization: better solutions at what cost? → Factoring changes the quantum conversation →

How it works

The dots on the grid lines are data qubits, and each square is a check on the four qubits around it; real chips measure each check with an extra qubit. An error (ember) flips a qubit. A check lights up when an odd number of its qubits have flipped, so it reports that something changed nearby, never the stored information itself. A decoder pairs up lit checks and flips the qubits along a path between them (the coloured lines); faint tracks are earlier fixes. A fix need not retrace the error: together they form a closed loop, which does no harm. Trouble starts only when errors and fixes together stretch right across the patch, which changes the stored information: a logical error.

Why it matters

You cannot copy an unknown quantum state to keep a backup, so error correction spreads one logical qubit across many physical ones and checks them like this, round after round. It only pays off when the hardware's error rate is below a threshold; then bigger codes make logical errors rarer. Google reported that behaviour for a stored logical qubit on its Willow chip in 2024. When a company counts qubits, ask how many are physical and how many are logical.

Try it

Switch the noise to high. Lit checks crowd together, pairings become ambiguous, and the decoder guesses wrong more often. This model is simplified: it has only bit-flip errors and perfect checks, while real codes also handle phase errors and faulty checks.

Go deeper: From physical qubits to logical qubits → A quantum memory below threshold →